Agent action generation on AgentGym and BrowseComp (test) benchmark leaderboard · SOTA2 Research

Agent action generation on AgentGym and BrowseComp (test)

75.52048 ScoreGPT-4o-mini

Evaluation Results

Method	Links
GPT-4o-mini2026.02		75.5	81	54.5	39.8	81.8	92	98.9	45.3	71.1
GPT-4o-mini2026.02		73.4	82	52.5	37.8	55	89.4	97.3	40.2	66
Qwen3-8B2026.02		73	91.5	70.3	54.9	83	68.4	87.5	51.1	72.5
GPT-4o-mini2026.02		72.9	62	62	45.5	52	87.2	96.8	38.4	64.6
GPT-4o-mini2026.02		72.9	71	56.2	42.2	66.7	91.2	98.9	40.7	67.5
Qwen3-8B2026.02		72.4	78	67.5	44.9	74.5	68.2	89.5	44.3	67.4
Qwen3-8B2026.02		71.6	87.5	71.5	52.2	75.2	69.4	89.9	49.1	70.8
Qwen3-8B2026.02		70.9	83	67.7	44.4	82.3	67.5	85.1	50.2	68.9
Qwen3-8B2026.02		70.6	78	71	46.5	80.5	64.6	88.9	50.4	68.8
Qwen3-8B2026.02		70.2	44	59.5	40.5	65.5	70	77.7	44.8	59
Llama3.1-8B2026.02		70.1	79.7	21.5	45.4	68.2	58.6	87	38	58.6
Llama3.1-8B2026.02		69.8	81.5	31.5	37.6	62.7	53.8	88.3	38	57.9
Llama3.1-8B2026.02		69.7	46.5	18.8	32.3	46.5	52.4	78.5	38.8	47.9
Llama3.1-8B2026.02		68.7	82.5	28.5	43.6	70	55.9	84.1	40.5	59.2
Llama3.1-8B2026.02		68.6	82.3	31.5	43.8	70	56.8	86.3	39.3	59.8
Llama3.1-8B2026.02		68.1	48	20.3	29.1	49.1	54	79.9	38.2	48.3
Llama3.1-8B2026.02		68	77.2	31.5	39.5	65	56.5	88.6	40.4	58.3
Llama3.1-8B2026.02		67.8	79	25.3	44.9	72	55.4	86.5	38.9	58.7
Qwen3-8B2026.02		66.9	74	68.2	40	71.3	67.5	85.3	45.8	64.9
Qwen3-8B2026.02		65.6	39	61	23.3	67.5	63.5	75.1	40.3	54.4